2

My mobile web app utilizes navigator.geolocation.watchPosition to track polylines and waypoints for historical walking tours. (A customer requirement no-native app)

My first testing was done in a vehicle due to bad weather and seemed to produce reasonable results when plotting the polyline on an embedded Google Map in real-time. However, when walking it produced a more zig zag line.

Some research has revealed that I need to filter the retrieved location instead of just calculating the distance of two points.

The suggested filter, Kalman filter, seems to be something over my head, and I'm wondering whether there is an open source implementation for JavaScript available, which can be of help for my use case.

An open source C/C++ implementation of the Kalman filter that could be compiled to JavaScript using emscripten might be of help too.

bardu
  • 727
  • 2
  • 8
  • 24

1 Answers1

3

I posted a simple Kalman filter that can be used for geolocation data in an answer to this question: Smooth GPS data . It was designed for Android, where the GPS location readings are accompanied by a single error-bar number called accuracy which is measured in metres. That means that for each GPS location reading, with 67% probability the true location is within the circle centered on the specified location, and the radius of the circle is given by the accuracy number. If your situation is different then a bit of work will be required, and if so I would suggest that you edit the question to spec out in more detail what inputs you have for the Kalman filter.

Community
  • 1
  • 1
Stochastically
  • 7,616
  • 5
  • 30
  • 58
  • Thanks for joining. I came across your simple Kalman filter before and planning to implement your Java class in JavaScript. Ones I have it done could we communicate about a test case to verify my implementation? Since the requirement is to use only HTML5 APIs I'm going to edit my original question as soon as I have all pieces together based on tests I have already done. – bardu Dec 17 '13 at 17:29
  • For a test case, one thing you could do would be to use your app to store real data (both location and the `accuracy` number) and then run that data through your implementation for different values of the `Q` parameter. You should see that `Q` controls the extent of the smoothing. – Stochastically Dec 18 '13 at 09:00